Skip to content

Commit

Permalink
Update README.md
Browse files Browse the repository at this point in the history
  • Loading branch information
Avic3nna authored Jun 11, 2024
1 parent 4ded644 commit 01cab87
Showing 1 changed file with 1 addition and 4 deletions.
5 changes: 1 addition & 4 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,10 +2,7 @@


> [!Important]
> STAMP v1.1.0 now uses PyTorchs FlashAttentionV2 implementation, which significantly improves memory efficiency when training, allowing you to use larger batch sizes and more tiles per patient. **Important:** The drawback is that you *cannot* deploy a saved model from STAMP version <= 1.0.3 with this or later versions. It is therefore recommended to only update to the latest version of STAMP when starting new experiments. Additionally, the optimizer has been updated from Adam to AdamW.
> [!Important]
> STAMP v1.0.3 now has built-in support for the [UNI Feature extractor](https://www.nature.com/articles/s41591-024-02857-3). Using it will require a Hugging Face account with granted access to the UNI model. For details on fair use, licensing and accessing the UNI model weights, refer to the [UNI GitHub repository](https://www.github.com/mahmoodlab/UNI.git). Note that the installation instructions within the [STAMP protocol paper](https://arxiv.org/abs/2312.10944v1) refer to v1.0.1 of the software, and that v1.0.3 has updated installation steps, see below. The README file will always contain the most up-to-date installation instructions.
> STAMP v1.1.0 now uses PyTorch's FlashAttentionV2 implementation, which substantially improves memory efficiency when training. With this update, is that you *cannot* deploy a saved model from STAMP version ≤ 1.0.3 with this or subsequent versions. Therefore, it is recommended to only update to the latest version of STAMP when starting new experiments. Additionally, the optimizer has been updated from Adam to AdamW. Lastly, STAMP has built-in support for the [UNI Feature extractor](https://www.nature.com/articles/s41591-024-02857-3). Using it will require a Hugging Face account with granted access to the UNI model. For details on fair use, licensing and accessing the UNI model weights, refer to the [UNI GitHub repository](https://www.github.com/mahmoodlab/UNI.git). Note that the installation instructions and results within the [STAMP protocol paper](https://arxiv.org/abs/2312.10944v1) refer to v1.0.3 of the software. The README file will always contain the most up-to-date installation instructions.
# STAMP protocol <img src="docs/STAMP_logo.svg" width="250px" align="right" />
A protocol for Solid Tumor Associative Modeling in Pathology. This repository contains the accompanying code for the steps described in the [preprint](https://arxiv.org/abs/2312.10944v1):
Expand Down

0 comments on commit 01cab87

Please sign in to comment.